AI FTP Detection

An interesting point here that always gets me is… do you need a long interval to push hard for a long time? Isn’t revolver pushing you hard for a long time? Is there any doubt that your long term performance has a huge impact on how much you can increase the targets on Revolver? Generally you can’t increase it past an IF of 1, ie past where NP is above FTP! I checked and even ignoring NP, the average power from revolver is about %80 of FTP, including the warmup. Now that’s certainly a lot easier than 100%, but it means 80% o the needed effort to wear you down is already coming from total energy expenditure during the time frame, even if it had been smoothed out. Of course using NP here to estimate FTP is less accurate because getting within that last 20% correctly depends on understanding your abilities (and recovery from) above threshold some (sort of your personalized NP correction, or 4DP parameters), but data on that is probably easier to find too, and is a correction. Standard NP already does a reasonable job usually.

A corollary to this is that just because there are a lot of short intervals, even ones where you would say you were working HARD, that still doesn’t mean you have your MAP or AC. You generally are not going all out on revolver intervals in the 1-rep-max sense. You might go all out on the first one, but you probably know better. There aren’t so many places in either indoor or outdoor training where you ever do that. And so estimating that also requires some disentanglement of FTP, and both require assumptions that enough of the efforts are truly pushed.

Suff used to have more FTP (or revolver-style NP/FTP) efforts though and less emphasis on erg mode They used to have instructions like "if you find you’re hitting 103% on many workouts, it’s probably time to bump up FTP, and that actually was pretty reasonable, no AI required. But with 100% efforts now encouraged only using the FF, and erg mode becoming the norm, that doesn’t work so well now. Why is it ok to tell people to push all out on FF every few weeks, but not on Thin Air? I have no idea, but that’s the new way.

As someone above said, at the end of the day, many of us just like to ride our bikes. If you just understand that most activities should not be at IF of 1 all the time, but also stop using ERG so much (I need to break the habit now myself) and let yourself go hard sometimes when you’re feeling spunky, hit that up arrow back up to 100%, and allow your legs to push on to 102% if you’re feeling it, you’ll naturally find when it’s time to up FTP. Sure you’ll find there are rides YOU can hit IF of 105 on some and some YOU can only hit 95, and that’s about your ability to hit the mix of intervals. So what? Do you even need 4DP to figure that out? Or maybe you just need to ride a lot, and sometimes ride hard?

Or just add it with the option of using it or not. Trainerroad users can still opt for the actual test if preferred.

OK, showing my ignorance of the topic but I think this is semantics; I used Xert for a good few years & it looked & felt like AI to me, in so far as it ‘learnt’ my rider characteristics giving it enough data over time.
So it is a ‘machine’ learning about me, and making recommendations & predictions based on past data, replacing human coach.
A dictionary definition of AI is:
“*Artificial intelligence is the simulation of human intelligence processes by machines, especially computer systems”.
Xert seems to work like that to me…

1 Like

I don’t think Xert is AI. It’s a strictly prescribed algorithm created by its designers. To be AI, it would need to be able to modify the algorithm based on experiments and results. IMO, a true AI trainer would likely dispense with FTP as being too simplistic (sorta like 4DP attempts). FTP is pretty arbitrary, a max power test for an hour. (many other tests try to estimate this while being easier to accomplish). Why not max power for 45 min, or 73 min or something else? There’s nothing physiologically special or fundamental about an hour. Long ago it was decided it would be convenient to divide one revolution of the Earth into 24 equal time intervals. That’s where an hour comes from.


With Xert you don’t do an ‘FTP’ test of any description. The reason I originally hoped onto this thread was to highlight this, and to note that although I had not done an ‘FTP’ test for some long time, my ‘20 min FTP’ figure in Xert was very, very close to the result I go from my 4DP test I took when joining Wahoo System a few months back. I appreciate that this is but one metric of an individuals power profile on a bike. So all I’m saying is that there are developments happening out there that could by utilised by Wahoo. BTW doing the 4DP test reminded me of how hard old style ‘FTP Tests’ are!!!

1 Like

Perfectly put. Similarly, I don’t worry too much that my Wattbike is probably overdue a service/recalibration - the numbers are what they are, so long as they read consistently they are doing their thing in harmony with SYSTM. I only pay attention to their trending along with perceived effort - ie. are they going up or down, am I’m finding workouts easier or harder - from which I can reasonably expect my outdoor performance/experience to similarly trend.

It’s very very hard to replicate indoor and outdoor training and performance 100%, so why bother trying - almost certainly only going lead to frustration, spending more time lining up numbers and compensating for various factors - all eating into energy otherwise spent on actually getting fitter! “Close enough” is good enough for me, also encourages the importance of mixing in both indoor AND outdoor training… which is no bad thing.

1 Like

Off topic: I assume you ignore it (as in decline modifying your set ftp), as in that’s why you’ve seen this prompt a number of times? I presumed Garmin only offered a newly detected FTP if it was an improvement on what it currently has set, crudely making a calculation based on seeing an improved 20 minute power average somewhere within the ride? Said differently, while it might be using some FirstBeat analytics, I didn’t think it’s all that sophisticated as in it can’t detect an ftp from sub-maximum efforts, and/or one lower than currently configured?

Getting back to the original question/suggestion;

Maybe a useful feature for SYSTM would be a AI/ML-based prompt suggesting that, based on analysis of recent workouts, the user might have made some noticeable fitness gains (or losses) and that it could be useful to do a FF/HM to find out, the result of which will (most importantly) adjust your future workouts accordingly.

This feature could be disabled (for those it doesn’t work for, eg. regularly prompting to retest yet with the same results, though arguably this could also point to a user struggling to execute the test to their full potential), would factor in time since last test and where abouts through a training program (eg. near the beginning = better to be working with accurate numbers, close to end = might as well see it out before a retest to review gains).

Furthering this, it would then place Wahoo Science in the perfect position to analyse comparisons between their “detections” and subsequent FF/HM retest results, providing evidence for what they may do with this feature moving forward - my guess/prediction would be it to be highly accurate in terms of “yes the vast majority of users prompted to retest did indeed come out with usefully different numbers” yet with less accuracy on predicting what these numbers would actually be across the board (in some cases good, in others terrible, as is the way with modelling and statistical analysis).


Yep. I’ve said a few time that I view FTP simply as a result of a 1 hour max power test. That’s it. There’s no underlying truth or reality that’s being quantified. Other so-called FTP tests provide their result which is then adjusted to approximate what an average athlete with that result might do in a 1 hr test, but there will be considerable differences between individuals. For the purposes of training, I stick with one test, try to do it in a consistent way and with a consistent lead-up to it. For SYSTM, that’s been the FF, and usually in the context of the 7 day Test Prep Plan that includes a HM on day 3.

1 Like

Yes! The marketing misuse of the term AI drives me crazy

1 Like

FWIW, I am working on a Masters in Data Science specifically for projects like this. I’d like to think I have the physiology/coaching side down. Just need a few more tools in the toolbox to integrate that knowledge into functional programs to give the best training advice to everyone who wants it. :slight_smile:


Wahoo Chat4DP 4TW :thinking:

1 Like